Evaluate Model on a Test Set

Hello, may i know if i can evaluate Yolov4, Ssd as well as Faster Rcnn TLT model on a test set? (not the validation set that is used during training)

I did read a forum where it states that user is able to evaluate detectnet model using a test set. Was wondering if the method to evaluate the faster rcnn model similar as well? here is the forum that i have read Calculate mAP of tlt using custom dataset - #5 by abrar.shahriar

Yes, you can.
Actually you can run
yolo_v4 evaluate xxx
ssd evaluate xxx


I do realize that when running the evaluate command, the model will evaluate on the validation set, but if i change the configuration file and specify the validation_data_sources to be the path of my test set and run the evaluate command, it will evaluate the model performance on the test set am i right?


How about for Faster Rcnn model?

Please set the path in dataset_config .

So basically, if i want to evaluate frcnn model on a test set, i have to convert my test set into tfrecords, and then sepcify the path of it to validation_data_source which is under dataset_config and also remove the line validation_fold = 0 am i right?

For detectnet_v2 , yes.
For ssd,dssd,yolo_v3, yolo_v4, they can be images/labels. Tfrecords is not a must.

ok will give it a shot, sorry for troubling you and thank you for your help :)


This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.